This study focuses on the complex flow channel application scenarios within Francis turbine units (FTUs), and proposes the physics-data dual-driven surrogate model for predicting the pressure pulsation characteristics of the FTUs flow channel. The aim is to address the challenges of insufficient generalization ability and lack of physical consistency in traditional data-driven models when dealing with limited samples. Initially, this paper integrates statistical anomaly detection and physical rules to conduct multi-layer cleaning of the measured operational data of the FTUs. Subsequently, physical laws are embedded into the physics-data dual-driven surrogate model through strong theoretical constraints and deviation corrections, ensuring that prediction results remain confined within physical boundaries. Ultimately, the multi-factor variable-weighted loss function is constructed to dynamically adjust the model in real time. This study compares and analyzes the proposed surrogate model with the traditional pure data-driven neural network (DDNN). Experimental results show that compared with the DDNN surrogate model, the proposed model has higher accuracy and robustness, and effectively ensures that the prediction results conform to physical laws, thereby providing strong support for the safe operation and optimization of the FTUs.